Hong-Stein Model

The Hong & Stein (1999) “Unified Theory of Underreaction, Momentum Trading, and Overreaction” explains the joint pattern of short-run momentum and long-run reversal through the interaction of two boundedly-rational investor types, rather than the biased beliefs of a single representative agent.

Setup. A single risky asset exists in fixed supply. Two types of agents trade it:

Newswatchers observe a private signal about fundamentals each period but do not condition on current or past prices. Information about the asset diffuses gradually across the population — at any given moment, each newswatcher sees only a fraction of the total information that has been released, and full diffusion takes $z$ periods.

Momentum traders observe past prices but not fundamentals. They are restricted to simple univariate trading strategies (cannot run multivariate regressions on price history) and condition only on the cumulative price change over the last $k$ periods.

Both agent types are otherwise rational subject to their information constraint. Newswatchers’ forecasts are rational given the subset of information they observe; momentum traders’ forecasts are constrained-optimal given that they can only use a univariate function of past prices.

Mechanism.

Newswatchers alone produce under-reaction. When a piece of fundamental news arrives, only a fraction of newswatchers see it immediately. Each period, the news diffuses to a new tranche of newswatchers, who then trade on it. Prices adjust gradually, generating $1$ – $z$ period continuation but never overshooting fundamental value. With only newswatchers, there is under-reaction and momentum but no long-term reversal.

Adding momentum traders mitigates under-reaction but creates eventual over-reaction. Momentum traders see early newswatcher trades push prices, infer that more news must be coming, and pile in. Their trades push prices further than the news justifies. The under-reaction is shortened, but at the cost of generating overshoot — and therefore long-run reversal — that would not exist with newswatchers alone.

Key insight: a negative externality. Each cohort of momentum traders trades on a price signal that is partially driven by previous momentum-trader flows, not by fundamental news. Because they cannot tell the two apart with their univariate strategies, late-arriving momentum traders end up “chasing previous rounds of momentum trade rather than recent private info.” They make money early in the cycle (when the price signal is news-driven) and lose money late in the cycle (when the signal is mostly other momentum traders’ flows). This externality is what generates the eventual reversal.

Distinctive empirical predictions.

Information environment effect: momentum is stronger in stocks where information diffuses more slowly. Hong, Lim & Stein (2000) confirm this — controlling for size, momentum is much stronger in low-analyst-coverage stocks, and the gap widens at longer horizons.

Trading volume effect: momentum should be associated with high trading volume (because it stems from heterogeneous information across newswatchers). Lee & Swaminathan (2000) confirm that momentum is stronger when accompanied by high turnover.

Counter-intuitive welfare result: adding more momentum traders to the market can reduce market efficiency, because they amplify the eventual overshoot.

Within the over- and under-reaction family, Hong-Stein is the heterogeneous-agent alternative to the single-investor BSV model. BSV gets the joint pattern from biased beliefs in one head; Hong-Stein gets it from rational-but-constrained behavior interacting across many heads.

Intuition

Hong & Stein’s starting move is methodological: rather than psych out the representative investor, hold each agent’s rationality fixed and let them differ in what they pay attention to. The simplest such split is between agents who see fundamentals but not prices and agents who see prices but not fundamentals. Neither type is biased — newswatchers’ forecasts are rational given the news they have, and momentum traders’ forecasts are constrained-optimal given the univariate-strategy restriction. The deviations from efficiency emerge from how the two groups interact, not from cognitive errors.

Underreaction comes from gradual information diffusion. Imagine a piece of good news that, if known to everyone, would justify a $10$ -point price increase. In the Hong-Stein world, only $1/ z$ of newswatchers see the news in any given period, so the first-period price moves only $10/ z$ points. Over the next $z$ periods, additional cohorts of newswatchers see the news and trade on it, pushing the price gradually toward $10$ . The gradual adjustment looks like momentum from the outside — past price changes predict future price changes — and it is genuine in the sense that no one in the market is leaving money on the table given their information.

Overreaction comes from momentum traders entering. Momentum traders see the price drifting up, assume “the trend means something,” and buy. Their buying pushes the price further. Critically, momentum traders cannot distinguish a price move that comes from new information arriving at later cohorts of newswatchers from a price move that comes from previous rounds of momentum trading. Their univariate strategy treats both as “trend.” So as the cycle continues, later cohorts of momentum traders end up trading on a signal that is dominated by earlier momentum traders’ flows, not by fundamental news. The price overshoots, and when newswatcher diffusion completes and the fundamental story is fully priced, the overshoot eventually reverses.

The counter-intuitive welfare result falls out cleanly: adding more momentum traders to a market with under-reaction does not push prices toward fundamentals — it pushes them past fundamentals. The momentum traders’ arbitrage is “constrained optimal” for them individually but creates a negative externality on the next generation of momentum traders. This is one of the cleanest examples in behavioral finance of how individually-rational behavior can aggregate into market-level inefficiency.

Properties

(Newswatchers Alone Yield Underreaction Without Reversal) In a market populated only by newswatchers, prices adjust gradually to news but never overshoot.

This is the model’s first stage. Under gradual information diffusion alone — each newswatcher sees a private signal at staggered times and trades immediately on the information she has — prices climb monotonically toward fundamental value over $z$ periods. The autocorrelation of returns is positive at horizons up to $z$ , generating a momentum signal, but there is no overshoot at any point and therefore no long-run reversal. This first stage isolates the underreaction mechanism: gradual diffusion is enough to create momentum but is not enough to create the joint continuation-and-reversal pattern that the data demand. Hong & Stein use this as the baseline against which the second stage (adding momentum traders) is measured.

(Momentum Traders Mitigate Underreaction at the Cost of Creating Overshoot) The cumulative impulse-response of price to news rises above $1$ before slowly returning to $1$ .

Hong & Stein’s Figure $2$ shows the cumulative price response to a single unit of fundamental news under different assumptions about momentum-trader risk tolerance $γ$ . With no momentum traders ( $γ = 0$ ), the impulse response rises smoothly to $1$ over $z$ periods and stays there. With momentum traders, the response rises faster (mitigating underreaction at short horizons) but overshoots $1$ , reaching a peak around period $12$ – $15$ , and then decays back toward $1$ over the next $20$ periods. The shape — fast climb, peak above $1$ , gradual decay — is the model’s central qualitative prediction and is what generates the joint continuation-then-reversal empirical pattern. Increasing $γ$ (more aggressive momentum traders) raises both the peak height and the eventual reversal magnitude.

(Negative Externality Across Momentum-Trader Generations) Momentum traders make money early in a news cycle and lose it late, but cannot tell which phase they are in.

This is the mechanism that drives the welfare result. Early in a cycle, the price drift is dominated by genuine fundamental news diffusing to additional newswatchers — momentum traders riding this drift earn positive returns. Late in a cycle, the price drift is dominated by previous momentum-trader flows being misread as further news by the next cohort — momentum traders riding this drift earn negative returns (the eventual reversal hits them). The univariate-strategy constraint means individual momentum traders cannot filter out which type of drift they are exploiting; they can only see total past price changes. Each momentum-trader cohort therefore inflicts a negative externality on the next, and the cycle ends in collective overshoot followed by collective reversal.

(Information-Environment Cross-Section) Stocks where information diffuses more slowly should display stronger and longer-lasting momentum.

This is Hong-Stein’s most distinctive empirical prediction and the cleanest test that distinguishes it from BSV. The diffusion parameter $z$ varies in the cross-section: information about a small, low-analyst-coverage stock takes longer to reach the trading population than information about a large, well-followed stock. Hong-Stein predicts that stocks with larger $z$ should have stronger momentum, and the prediction was confirmed out of sample by Hong, Lim & Stein (2000): controlling for size, momentum strategies earn roughly twice as much in low-analyst-coverage stocks as in high-analyst-coverage stocks, and the gap widens monotonically at longer holding horizons. BSV makes no such cross-sectional prediction because it has only one investor and one information source.

(Trading Volume Cross-Section) Momentum should be associated with high trading volume; high-volume momentum should be the most pronounced.

In Hong-Stein, momentum exists because newswatchers have heterogeneous private information at any given time, so they are trading with each other as the news diffuses. Trading volume is therefore mechanically tied to the information-diffusion process that generates the momentum. Lee & Swaminathan (2000) confirm this prediction: momentum effects are substantially stronger in high-volume subsamples than in low-volume subsamples. Again, BSV is silent on volume because its single representative agent has no one to trade with; the volume prediction is unique to heterogeneous-agent models like Hong-Stein.

(More Momentum Traders Can Reduce Efficiency) Adding momentum-trader capital to the market can move prices further from fundamentals, not closer.

This is the most counter-intuitive result in the paper, and it’s a direct consequence of the externality. In a standard arbitrage story, more arbitrageurs push prices toward fundamentals; in Hong-Stein, the momentum-trader “arbitrage” of underreaction creates overshoot, and adding more momentum traders (raising $γ$ ) makes the overshoot larger. The market becomes more efficient at short horizons (less under-reaction) but less efficient at medium horizons (more over-reaction). The implication for policy is that not all “smart money” pushes markets toward efficiency — when the smart money’s strategy is constrained to be univariate, more of it can hurt rather than help.

Examples

The two-cohort cycle, walked through

Imagine a single piece of fundamental news at $t = 0$ that, if priced rationally and instantly, would push the asset price from $100$ to $110$ . The diffusion parameter is $z = 12$ (information takes $12$ periods to fully diffuse to all newswatchers), the momentum-trader horizon is $j = 12$ (they look at the cumulative change over the last $12$ periods), and risk tolerance $γ$ is moderate.

Periods $1$ – $4$ (early cycle). Each period, $1/12$ of newswatchers see the news and buy. Without momentum traders, the price would climb linearly from $100$ to about $103.3$ by period $4$ . With momentum traders, the price climbs faster — say to $105$ — because momentum traders see the early drift and add buying pressure. They are correctly inferring that more news is coming and earning a positive return on the trade.

Periods $5$ – $12$ (mid cycle). The drift continues. Momentum traders keep buying. By period $12$ , all newswatchers have seen the news, so the fundamental-driven component of the price is now $110$ . But momentum traders have continued buying past this point, pushing the price to maybe $115$ . This is the overshoot — the price is now $5$ above fundamentals.

Periods $13$ – $24$ (late cycle). The newswatcher-driven component is done diffusing — there is no new fundamental information left to incorporate. But momentum traders entering at period $13$ still see a positive trailing $12$ -period return and keep buying. They cannot tell that the recent drift has been pure momentum-trader flow, not news. As the next cohort gradually realizes the news has run out, buying slows and the price starts drifting down. Eventually it returns to $110$ — but the momentum traders who bought at $115$ in periods $11$ – $13$ take losses on the way down.

Aggregating across many such cycles, the model produces autocorrelation that is positive at horizons up to $\sim z$ and negative at horizons beyond — exactly the empirical pattern of $1$ – $12$ month continuation followed by $13$ – $36$ month reversal.

The Hong-Lim-Stein cross-section

Hong, Lim & Stein (2000) test the model’s distinctive prediction that momentum should be stronger where information diffuses more slowly. They sort U.S. stocks into terciles by residual analyst coverage (analyst coverage minus the part predicted by firm size, so the residual captures coverage variation that is not just a size effect). Within each size group, they then run standard $(6, 6)$ momentum strategies and compare returns across coverage terciles.

The results match Hong-Stein cleanly. In the lowest-residual-coverage tercile (SUB $1$ ), the cumulative momentum-strategy return rises to about $20%$ over $36$ months and stays elevated. In the highest-residual-coverage tercile (SUB $3$ ), it rises only to about $10%$ and plateaus earlier. Both are positive — momentum exists in both groups — but the magnitude is roughly double in the low-coverage subsample, and the divergence is largest at longer horizons. The interaction with size is also as predicted: coverage matters more at the margin for small stocks, where the absolute number of analysts is low and an additional analyst makes a larger relative difference.

This is a sharp empirical victory for the gradual-information-diffusion mechanism. BSV cannot generate this prediction because it has only public earnings news and one investor — there is no analog of “information environment” in the BSV setup.

A constrained univariate momentum trader, formally

The Hong-Stein momentum trader observes the cumulative log price change over the last $k$ periods, $Δ P_{t - k : t}$ , and trades a position $x_{t} = γ \cdot Δ P_{t - k : t}$ . That is the entire strategy — a constant $γ$ times the trailing return.

Why is this “constrained-optimal”? Because Hong-Stein restricts momentum traders to univariate functions of past prices. They cannot, for instance, run a regression of returns on multiple lagged-return windows and weight them differently. The univariate restriction is what makes them unable to filter out which part of the recent price drift came from news vs. previous momentum-trader flow. If momentum traders were allowed multivariate strategies, they could in principle estimate the diffusion parameters and back out the news-driven vs. flow-driven components — and the overshoot would not arise. The bounded-rationality assumption isn’t that momentum traders are stupid; it’s that their strategy space is one-dimensional, and that one dimension is not enough to disentangle the two sources of past returns.

Diagram

The figure below sketches Hong & Stein’s central qualitative prediction (Figure $2$ in the paper): the cumulative price response to a single unit of fundamental news, with and without momentum traders.

\usepackage{tikz}
\begin{document}
\begin{tikzpicture}[x=0.32cm, y=2.6cm, >=stealth]
  % axes
  \draw[->,thick] (-1,0) -- (35,0) node[right] {lag (periods)};
  \draw[->,thick] (0,-0.05) -- (0,1.6) node[above] {cumulative price response};
  \foreach \x/\lab in {0/0, 6/6, 12/12, 18/18, 24/24, 30/30} {
    \draw (\x,0.02) -- (\x,-0.02) node[below] {\lab};
  }
  \foreach \y/\lab in {0.5/0.5, 1.0/1.0, 1.4/1.4} {
    \draw (-0.3,\y) -- (0.3,\y) node[left] at (-0.3,\y) {\lab};
  }
  % baseline (newswatchers only): linear ramp to 1 over 12 periods, then flat
  \draw[thick, blue] (0,0) -- (12,1.0) -- (34,1.0);
  \node[blue, above right] at (15.5,0.99) {newswatchers only};
 
  % with momentum traders: faster early climb, peak ~1.35 at lag 13, slow decay back to 1
  \draw[thick, red] plot[smooth] coordinates {
    (0,0) (2,0.25) (4,0.5) (6,0.75) (8,0.95) (10,1.15) (12,1.30) (13,1.35)
    (15,1.32) (18,1.22) (21,1.13) (24,1.07) (28,1.03) (32,1.01) (34,1.00)
  };
  \node[red, above] at (13,1.42) {with momentum traders};
 
  % overshoot annotation
  \draw[<->, gray] (13,1.0) -- (13,1.35);
  \node[gray, right] at (13.2,1.18) {overshoot};
 
  % reversal annotation
  \draw[->, gray, thick] (16,1.30) to[bend right=15] (28,1.05);
  \node[gray] at (24,1.27) {reversal};
\end{tikzpicture}
\end{document}

The newswatcher-only impulse response (blue) climbs linearly to $1.0$ over $z = 12$ periods and stays there: under-reaction at short horizons, no overshoot, no reversal. Adding momentum traders (red) shifts the climb earlier and steeper, then overshoots the long-run value of $1.0$ around lag $13$ , then decays slowly back over the next $20$ periods. The gap above $1.0$ is the overshoot, and the slow decay back is the long-run reversal that BSV-style risk-adjusted return regressions detect at $13$ – $36$ month horizons.

tags	domain/finance, type/concept, level/intermediate, status/complete
aliases	Hong & Stein 1999, Newswatchers and Momentum Traders, Unified Theory of Underreaction Momentum and Overreaction

Project Alexandria

Explorer

Hong-Stein Model

Intuition

Properties

(Newswatchers Alone Yield Underreaction Without Reversal) In a market populated only by newswatchers, prices adjust gradually to news but never overshoot.

(Momentum Traders Mitigate Underreaction at the Cost of Creating Overshoot) The cumulative impulse-response of price to news rises above $1$ before slowly returning to $1$ .

(Negative Externality Across Momentum-Trader Generations) Momentum traders make money early in a news cycle and lose it late, but cannot tell which phase they are in.

(Information-Environment Cross-Section) Stocks where information diffuses more slowly should display stronger and longer-lasting momentum.

(Trading Volume Cross-Section) Momentum should be associated with high trading volume; high-volume momentum should be the most pronounced.

(More Momentum Traders Can Reduce Efficiency) Adding momentum-trader capital to the market can move prices further from fundamentals, not closer.

Examples

Diagram

See Also

Graph View

Table of Contents

Backlinks

Project Alexandria

Explorer

Hong-Stein Model

Intuition

Properties

(Newswatchers Alone Yield Underreaction Without Reversal) In a market populated only by newswatchers, prices adjust gradually to news but never overshoot.

(Momentum Traders Mitigate Underreaction at the Cost of Creating Overshoot) The cumulative impulse-response of price to news rises above 1 before slowly returning to 1.

(Negative Externality Across Momentum-Trader Generations) Momentum traders make money early in a news cycle and lose it late, but cannot tell which phase they are in.

(Information-Environment Cross-Section) Stocks where information diffuses more slowly should display stronger and longer-lasting momentum.

(Trading Volume Cross-Section) Momentum should be associated with high trading volume; high-volume momentum should be the most pronounced.

(More Momentum Traders Can Reduce Efficiency) Adding momentum-trader capital to the market can move prices further from fundamentals, not closer.

Examples

Diagram

See Also

Graph View

Table of Contents

Backlinks

(Momentum Traders Mitigate Underreaction at the Cost of Creating Overshoot) The cumulative impulse-response of price to news rises above $1$ before slowly returning to $1$ .